Welcome to the Common Component Architecture Tutorial

1. Welcome to theCommon Component Architecture Tutorial ACTS Collection Workshop 20 August 2010

2. Supplementary material for handouts About the Printed Notes • The printed (or PDF) version of these presentations includes additional slides marked “Supplementary material for handouts” • Additional material to address questions sometimes raised, or provide more detail on a topic • We are happy to discuss this material if asked

3. Introductions • David E. Bernholdt (Oak Ridge National Laboratory) • Dietmar Ebner (Lawrence Livermore National Laboratory) • Wael R. Elwasif (Oak Ridge National Laboratory) • Sameer Shende (U. Oregon)

4. Who We Are: The Common Component Architecture (CCA) Forum • Combination of standards body and user group for the CCA • Define specifications for high-performance scientific components & frameworks • Promote and facilitate development of tools for component-based software development, components, and component applications • Open membership, quarterly meetings… General mailing list: cca-forum@cca-forum.org Web: http://www.cca-forum.org/

5. Agenda & Table of Contents These notes are available online at: http://www.cca-forum.org/tutorials/archives/2010/tutorial-2010-08-20

6. What You Should Take Away from this Tutorial • The concepts of component-based software development (CBSD) • Help organize, design, and manage large, complex software systems • The Common Component Architecture (CCA) is a CBSD approach specifically designed for high-performance computational science and engineering • The CCA is not the only way to do CBSD • Hands-on experience using and creating simple components using the CCA tools

7. Introduction to HPC ComponentSoftware

8. Managing Code Complexity Some Common Situations: • Your code is so large and complex it has become fragile and hard to keep running • You have a simple code, and you want to extend its capabilities – rationally • You want to develop a computational “toolkit” • Many modules that can be assembled in different ways to perform different scientific calculations • Gives users w/o programming experience access to a flexible tool for simulation • Gives users w/o HPC experience access to HPC-ready software How CCA Can Help: • Components help you think about software in manageable chunks that interact only in well-defined ways • Components provide a “plug-and-play” environment that allows easy, flexible application assembly

9. A toolkit to perform simulations of unsteady flames Solve the Navier-Stokes with detailed chemistry Various mechanisms up to ~50 species, 300 reactions Structured adaptive mesh refinement CFRFS today: 61 components 7 external libraries 9 contributors Example: Computational Facility for Reacting Flow Science (CFRFS) “Wiring diagram” for a typical CFRFS simulation, utilizing 12 components. CCA tools used: Ccaffeine, and ccafe-gui Languages: C, C++, F77

10. Helping Groups Work with Software Some Common Situations: • Many (geographically distributed) developers creating a large software system • Hard to coordinate, different parts of the software don’t work together as required • Groups of developers with different specialties • Forming communities to standardize interfaces or share code How CCA Can Help: • Components are natural units for • Expressing software architecture • Individuals or small groups to develop • Encapsulating particular expertise • Some component models (including CCA) provide tools to help you think about the interface separately from the implementation

11. Integrated state-of-the-art optimization technology into two quantum chemistry packages to explore effectiveness in chemistry applications Geographically distributed expertise: California - chemistry Illinois - optimization Washington – chemistry, parallel data management Effective collaboration with minimal face-to-face interaction Schematic of CCA-based molecular structure determination quantum chemistry application. Components based on: MPQC, NWChem (quantum chem.), TAO (optimization), Global Arrays, PETSc (parallel linear algebra) CCA tools used: Babel, Ccaffeine, and ccafe-gui Languages: C, C++, F77, Python Example: Quantum Chemistry

12. Language Interoperability Some Common Situations: • Need to use existing code or libraries written in multiple languages in the same application? • Want to allow others to access your library from multiple languages? • Technical or sociological reasons for wanting to use multiple languages in your application? How CCA Can Help: • Some component models (including CCA) allow transparent mixing of languages • Babel (CCA’s language interop. tool) can be used separately from other component concepts

13. hypre High performance preconditioners and linear solvers Library written in C Babel-generated object-oriented interfaces provided in C, C++, Fortran LAPACK07 Update to LAPACK linear algebra library To be released 2007 Library written in F77, F95 Will use Babel-generated interfaces for: C, C++, F77, F95, Java, Python Possibly also ScaLAPACK (distributed version) Examples “I implemented a Babel-based interface for the hypre library of linear equation solvers. The Babel interface was straightforward to write and gave us interfaces to several languages for less effort than it would take to interface to a single language.” -- Jeff Painter, LLNL. 2 June 2003 CCA tools used: Babel, Chasm

14. Coupling Codes Some Common Situations: • Your application makes use of numerous third-party libraries • Some of which interact (version dependencies) • You want to develop a simulation in which your code is coupled with others • They weren’t designed with this coupling in mind • They must remain usable separately too • They are all under continual development, individually • They’re all parallel and need to exchange data frequently How CCA Can Help: • Components are isolated from one another • Interactions via well-defined interfaces • An application can include multiple versions of a component • Components can be composed flexibly, hierarchically • Standalone application as one assembly, coupled simulation as another • CCA can be used in SPMD, MPMD, and distributed styles of parallel computing • CCA is developing technology to facilitate data and functional coupling of parallel applications

15. Proof-of-principle of using CCA for integrated whole-device modeling needed for the ITER fusion reactor Couples radio frequency (RF) heating of plasma with transport modeling Coarse-grain encapsulation of pre-existing programs Follow-on plans for RF, transport, and magneto-hydrodynamics “Wiring diagram” for integrated fusion simulation. Components based on: AORSA, Houlberg’s transport library New components: Driver, State CCA tools used: Babel, Chasm, Ccaffeine, ccafe-gui Languages: C++, F90, Python Example: Integrated Fusion Simulation

16. What are Components? • No universally accepted definition in computer science research, but key features include… • A unit of software development/deployment/reuse • i.e. has interesting functionality • Ideally, functionality someone else might be able to (re)use • Can be developed independently of other components • Interacts with the outside world only through well-defined interfaces • Implementation is opaque to the outside world • Can be composed with other components • “Plug and play” model to build applications • Composition based on interfaces

17. What is a Component Architecture? • A set of standardsthat allows: • Multiple groups to write units of software (components)… • And have confidence that their components will work with other components written in the same architecture • These standards define… • The rights and responsibilities of a component • How components express their interfaces • The environment in which components are composed to form an application and executed (framework) • The rights and responsibilities of the framework

18. Interoperable components (provide same interfaces) • A Simple Example: Numerical Integration Components FunctionPort IntegratorPort FunctionPort NonlinearFunction FunctionPort MidpointIntegrator GoPort IntegratorPort LinearFunction FunctionPort Driver IntegratorPort FunctionPort PiFunction RandomGeneratorPort RandomGeneratorPort MonteCarloIntegrator RandomGenerator

19. An Application Built from the Provided Components FunctionPort IntegratorPort FunctionPort NonlinearFunction FunctionPort MidpointIntegrator GoPort IntegratorPort LinearFunction FunctionPort Driver IntegratorPort FunctionPort PiFunction Hides compexity: Driver doesn’t care that MonteCarloIntegrator needs a random number generator RandomGeneratorPort RandomGeneratorPort MonteCarloIntegrator RandomGenerator

20. Another Application… FunctionPort IntegratorPort FunctionPort NonlinearFunction FunctionPort MidpointIntegrator GoPort IntegratorPort LinearFunction FunctionPort Driver IntegratorPort FunctionPort PiFunction RandomGeneratorPort RandomGeneratorPort MonteCarloIntegrator RandomGenerator

21. And Many More… Dashed lines indicate alternate connections FunctionPort IntegratorPort FunctionPort NonlinearFunction FunctionPort MidpointIntegrator GoPort IntegratorPort LinearFunction FunctionPort Driver IntegratorPort FunctionPort PiFunction RandomGeneratorPort RandomGeneratorPort MonteCarloIntegrator Create different applications in "plug-and-play" fashion RandomGenerator

22. Relationships: Components, Objects, and Libraries • Components are typically discussed as objectsor collections of objects • Interfacesgenerally designed in OOterms, but… • Component internals need not be OO • OO languages are not required • Component environments can enforce the use of published interfaces (prevent access to internals) • Libraries can not • It is possible to load several instances (versions) of a component in a single application • Impossible with libraries • Components must include some code to interface with the framework/component environment • Libraries and objects do not

23. What is the CCA? • Component-based software engineering has been developed in other areas of computing • Especially business and internet • Examples: CORBA Component Model, COM, Enterprise JavaBeans • Many of the needs are similar to those in HPC scientific computing • But scientific computing imposes special requirements not common elsewhere • CCA is a component environment specially designed to meet the needs of HPC scientific computing

24. Special Needs of Scientific HPC • Support for legacy software • How much change required for component environment? • Performance is important • What overheads are imposed by the component environment? • Both parallel and distributed computing are important • What approaches does the component model support? • What constraints are imposed? • What are the performance costs? • Support for languages, data types, and platforms • Fortran? • Complex numbers? Arrays? (as first-class objects) • Is it available on my parallel computer?

25. CCA: Concept and Practice • In the following slides, we explain important concepts of component-based software from the CCA perspective • We also sketch how these concepts are manifested in code (full details in the Hands-On) • The CCA Specification is the mapping between concept and code • A standard established by the CCA Forum • Expressed in the Scientific Interface Definition Language (SIDL) for language neutrality (syntax similar to Java) • SIDL can be translated into bindings for specific programming languages using, e.g., the Babel language interoperability tool

26. IntegratorPort FunctionPort FunctionPort MidpointIntegrator NonlinearFunction CCA Concepts: Components • A component encapsulates some computational functionality • Components provide/use one or more interfaces • A component with no interfaces is formally okay, but isn’t very interesting or useful • In SIDL, a component is a class that implements (inherits from) gov.cca.Component • This means it must implement the setServices method to tell framework what ports this component will provide and use • gov.cca.Componentis defined in the CCA specification

27. IntegratorPort FunctionPort FunctionPort MidpointIntegrator NonlinearFunction CCA Concepts: Ports • Components interact through well-defined interfaces, or ports • A port expresses some computational functionality • In Fortran, a port is a bunch of subroutines or a module • In OO languages, a port is an abstract class or interface • Ports and connections between them are a procedural (caller/callee) relationship, not dataflow! • e.g., FunctionPortcould contain a method like evaluate(in Arg, out Result) with data flowing both ways

28. IntegratorPort FunctionPort FunctionPort MidpointIntegrator NonlinearFunction CCA Concepts: Provides and Uses Ports • Components may provide ports – implement the class or subroutines of the port ( ) • Providing a port implies certain inheritance relationships between the component and the abstract definition of the interface (more details shortly) • A component can provide multiple ports • Different “views” of the same functionality, or • Related pieces of functionality • Components may use ports – call methods or subroutines in the port ( ) • Use of ports is just like calling a method normally except for a little additional work due to the “componentness” (more details shortly) • No inheritance relationship implied between caller and callee • A component can use multiple ports “Provides” Port “Uses” Port

29. IntegratorPort FunctionPort MidpointIntegrator Key: SIDL inheritance keywords = Inheritance • • • Components and Ports (in SIDL) package gov.cca { interface Component { void setServices(…); } } package gov.cca { interface Port { } } package integrators { interface IntegratorPort { double integrate(…}; } } extendsgov.cca.Port package integrators { class Midpoint { } } implements gov.cca.Component, integrator.IntegratorPort double integrate(…); void setServices(…);

30. IntegratorPort FunctionPort FunctionPort MidpointIntegrator NonlinearFunction Using Ports • Calling methods on a port you use requires that you first obtain a “handle” for the port • Done by invoking getPort() on the user’s gov.cca.Services object • Free up handle by invoking releasePort() when done with port • Best practice is to bracket actual port usage as closely as possible without using getPort(), releasePort() too frequently • Can be expensive operations, especially in distributed computing contexts • Performance is in tension with dynamism • can’t “re-wire” a ports that is “in use”

31. CCA Concepts: Frameworks • The framework provides the means to “hold” components and compose them into applications • Frameworks allow connection of ports without exposing component implementation details • Frameworks provide a small set of standard services to components • Framework services are CCA ports, just like on components • Additional (non-standard) services can also be offered • Components can register ports as services using the ServiceProvider port • Currently: specific frameworks are specialized for specific computing models (parallel, distributed, etc.) • Future: better integration and interoperability of frameworks

32. IntegratorPort FunctionPort FunctionPort MidpointIntegrator NonlinearFunction Components Must Keep Frameworks Informed • Components must tell the framework about the ports they are providing and using • Framework will not allow connections to ports it isn’t aware of • Register them using methods on the component’s gov.cca.Services object • addProvidesPort() and removeProvidesPort() • registerUsesPort() and unregisterUsesPort() • All are defined in the CCA specification • Ports are usually registered in the component’s setServices() method • Can also be added/removed dynamically during execution

33. Scientific software is increasingly diverse in use of programming languages In a component environment, users should not care what language a component is implemented in “Point-to-point” solutions to language interoperability are not suitable for a component environment The Babel language interoperability tool provides a common solution for all supported languages Scientific Interface Definition Language provides language-neutral way of expressing interfaces f77 f77 C f90/95 C f90 Python C++ C++ Python Java Java CCA Concepts: Language Interoperability More on Babel later!

34. Single component multiple data (SCMD) model is component analog of widely used SPMD model P0 P1 P2 P3 • • • • CCA Supports Parallelism -- by “Staying Out of the Way” of it • Each process loaded with the same set of components wired the same way • Different components in same process “talk to each” other via ports and the framework Components: Blue, Green, Red Framework: Gray • Same component in different processes talk to each other through their favorite communications layer (i.e. MPI, PVM, GA) Any parallel programming environments that can be mixed outside of CCA can be mixed inside

35. The performance of your application is as important to us as it is to you The CCA is designed to provide maximum performance But the best we can do is to make your code perform no worse, unless we give easy access to new algorithms. Facts: Measured overheads per function call arelow Most overheads easily amortized by doing enough work per call Other changes made during componentization may also have performance impacts Awarenessof costs of abstraction and language interoperability facilitates design for high performance Maintaining HPC Performance More about performance in notes

36. Performance Monitoring with TAU Before: Component1 Component2 After: Component2 Proxy for Component2 Component1 MasterMind(database) TAU(measure-ment) The Proxy Component Pattern • A “proxy” component can be inserted between the user and provider of a port without either being aware of it (non-invasive) • Proxy can observe or act on all invocations of the interface • Similar to aspect-oriented programming • For many purposes, proxies can be generated automatically from SIDL definition of the port Sample uses for proxy components: • Performance: instrumentation of method calls • Debugging: execution tracing, watching data values • Testing: Capture/replay

37. Additional material in notes Component Lifecycle • Composition Phase (assembling application) • Component is instantiatedin framework • Component interfaces are connected appropriately • Execution Phase (running application) • Code in components uses functions provided by another component • Decomposition Phase (termination of application) • Connections between component interfaces may be broken • Component may be destroyed In an application, individual components may be in different phases at different times Steps may be under human or software control

38. Is CCA for You? • Much of what CCA does can be done without such tools if you have sufficient discipline • The larger a group, the harder it becomes to impose the necessary discipline • Projects may use different aspects of the CCA • CCA is not monolithic – use what you need • Few projects use all features of the CCA… initially • Evaluate what your project needs against CCA’s capabilities • Other groups’ criteria probably differ from yours • CCA continues to evolve, so earlier evaluations may be out of date • Evaluate CCA against other ways of obtaining the desired capabilities • Suggested starting point: • CCA tutorial “hands-on” exercises

39. Take an Evolutionary Approach • The CCA is designed to allow selective use and incremental adoption • “SIDLize” interfaces incrementally • Start with essential interfaces • Remember, only externally exposed interfaces need to be Babelized • Componentize at successively finer granularities • Start with whole application as one component • Basic feel for components without “ripping apart” your app. • Subdivide into finer-grain components as appropriate • Code reuse opportunities • Plans for code evolution

40. The Primary Tools

41. The tools • Bocca – project environment • Ccaffeine – framework • SIDL – interoperability language • Babel – HPC language binding generator • CCA – specification for components, frameworks

42. CCA/Frameworks Component A Component B Babel Chasm Tools Module Overview • CCA Development Environment • Frameworks • Language interoperability tools CCA IDE

43. Bocca Development Environment • Provides a text-based, portable environment • Create or import SIDL and CCA based codes. • Automatic build system maintenance. • Easy to adopt or abandon while preserving code, build. • No GUI required. • Basis for common CCA toolkit installation. • Manages components in all Babel-supported languages (C, C++, Fortran, Java, Python).

44. Bocca Creates Skeletons for CCA • Including ports and interfaces • Give the SIDL name and an empty port or interface is created. • Including components and classes • Give the name and an empty component or class is created. • Some extra options: the component uses/provides ports, implemented interfaces or extended classes • Including build system • For all ports/components in the project • Implemented in any CCA supported language • Create applications with Ccaffeine GUI (today) • Including application composition (coming soon)

45. Bocca Example # create an empty but buildable CCA skeleton bocca create project myproj cd myproj ./configure bocca create port myJob bocca create component myWorker –-provides=myJob:job1 # fill in public functionality bocca edit port myJob # fill in implementation bocca edit component –i myWorker # compile application make

46. CCA/Frameworks Component A Component B Babel Chasm Tools Module Overview • CCA Interactive Development Environment • Frameworks • Language interoperability tools CCA IDE

47. Supports SIDL/Babel components Conforms to latest CCA specification (0.8) Also supports legacy CCA specification (0.5) Any C++ allowed with C and Fortran by C++ wrappers Provides command-line and GUI for composition Scripting supports batch mode for SPMD MPMD/SPMD custom drivers in any Babel language Ccaffeine is a Direct-Connect, Parallel-Friendly Framework Supported on Linux, AIX, OSX and is portable to modern UNIXes.

48. connect Driver IntegratorPort MonteCarloIntegrator IntegratorPortconnect Driver IntegratorPort MonteCarloIntegrator IntegratorPort connect MonteCarloIntegrator FunctionPort LinearFunction FunctionPort … User Connects Ports • Can only connect uses & provides • Not uses/uses • Not provides/provides • Ports connected by type not name • Port names must beunique within a component • Types must match across components • Framework puts info about provider of portinto using component’s Services object

49. CCA/Frameworks Component A Component B Babel Chasm Tools Module Overview • CCA Interactive Development Environment • Frameworks • Language interoperability tools CCA IDE

50. Programming language-neutral interface descriptions Native support for basic scientific data types Complex numbers Multi-dimensional, multi-strided arrays Automatic object-oriented wrapper generation Usable standalone or in CCA environment f77 f77 C C f90 f90 Babel vs. Python C++ C++ Python Java Java SIDL Facilitates ScientificProgramming Language Interoperability Supported on Linux, AIX, works on OSX, catamount; C (ANSI C), C++ (GCC), F77 (g77, Sun f77), F90 (Intel, Lahey, GNU, Absoft, PGI), Java (1.4)